A hybrid model for short-term dissolved oxygen content prediction

Abstract

Dissolved oxygen (DO) content is a significant indicator of water quality in intensive aquaculture ponds and is profoundly related to healthy fish growth. Accurately forecasting DO and analysing its change trends in aquaculture ponds is crucial for fish survival. A novel forecasting model based on a combination of complete ensemble empirical mode decomposition with an adaptive noise Lempel-Ziv complex (CEEMDAN-LZC) and a gated recurrent unit (GRU) with a generalized opposition-based learning particle swarm optimization algorithm (GOBLPSO) has been proposed to improve the prediction precision of DO. The DO content numerical sequence was decomposed and reconstructed into several new features by the CEEMDAN-LZC method in the modelling process. Independent models were structured to fit the components obtained above using GRUs, and the prediction value was superimposed to obtain the ultimate result. Furthermore, the time-step parameter t and the unit number parameter u, which significantly influence the performance of the GRU, were selected by the GOBLPSO algorithm. The simulation results based on DO content data in river crab culture ponds show that the proposed model has excellent essential feature extraction capability and is very powerful and reliable for DO content prediction in aquaculture.